Products & Tools

Google Will Let You Stop Gemini Training Without Losing Chats

Google will split Gemini's Keep Activity setting in two, letting users opt out of model training within weeks while keeping their full chat history for the first time.

Google is making it easier to opt out of Gemini AI model training – here’s how
Google is making it easier to opt out of Gemini AI model training – here’s howPexels / pixabay
By Sophie Lindqvist4 min read

Why it matters

  • Google will let Gemini users opt out of model training without deleting chat history within the next few weeks.
  • The single Keep Activity setting is being replaced by two independent settings for history retention and training.
  • Google was the only major AI maker requiring history deletion to opt out; ChatGPT, Copilot and Claude all allow opt-out with history kept.
  • Claude's AI training option is off by default; ChatGPT's is called "Improve the model for everyone."
  • Ziff Davis, ZDNET's parent, sued OpenAI in April 2025 over alleged copyright infringement in AI training.

Google will let Gemini users opt out of AI model training without deleting their chat history — a change the company says will roll out over the next few weeks. That ends an all-or-nothing trade-off that has set Google apart from its rivals, and not in a good way.

Until now, anyone uncomfortable with Google using their Gemini conversations to improve its AI models had exactly one way out: wipe the entire chat history. There was no middle ground. You either let Google train on your chats, or you lost them.

What is actually changing?

The current system runs through a single setting called Keep Activity. It controls two things at once: whether your chats get saved, and for how long — and whether Google can use them to train its models. The problem is obvious once you see it. You can't separate saving from training.

Google's fix is to split Keep Activity into two independent settings, according to an announcement the company shared with Gemini users by email.

  • History setting: keeps or deletes your past chats, and lets you choose how long your history stays saved.
  • Training setting: decides whether your chats — including all text and attachments — get used for model training.

"With this change, you will be able to keep your history while independently controlling generative AI model training," Google said in its email to users.

That single sentence carries the whole story. For the first time, the two decisions live in separate places. You can keep a year of conversations and still tell Google not to learn from them.

The change is scheduled to arrive "in the coming weeks," which means Gemini users should watch their settings menu rather than mark a calendar date.

Why does this matter?

It matters because Google has been the outlier. Nearly every major AI assistant offers a way to opt out of model training, and in almost every case you keep your conversation history when you use it.

Google didn't work that way. As ZDNET, which first reported the details, points out, Google has been the only major AI player to impose a restriction on people who wanted out of AI training. Opting out meant sacrificing your history — effectively a penalty for caring about privacy.

The company has also built a reputation over the years for being privacy-unfriendly, a track record that made this particular design choice feel less like an oversight and more like a pattern. So the split is a small step, but a meaningful one: it aligns Gemini with what users of other assistants already take for granted.

There's a broader context here too. Training data is the fuel of AI models, and the disputes around it are escalating. Ziff Davis, ZDNET's parent company, filed a lawsuit against OpenAI in April 2025 alleging it infringed Ziff Davis copyrights in training and operating its AI systems. Cases like that put pressure on every AI company to be clearer about what data they collect and how it gets used. Giving users a clean, independent off switch is part of that reckoning.

How do other AI assistants handle it?

Almost every AI assistant includes a model-training setting, and in most cases you have to find it and disable it yourself. But none of the others make you delete your history to use it.

Here's how the major players compare, based on ZDNET's rundown:

  • ChatGPT offers a setting called "Improve the model for everyone" — which is, as ZDNET puts it, double-speak for model training. It's easy enough to turn off, and your history stays.
  • Microsoft Copilot provides two toggles: "Training on conversation activity" and "Training on voice conversations." Disable both and you've opted out, chats intact.
  • Claude takes the friendliest approach: its AI training option is off by default. Users who want to check can look at the setting labeled "Help improve our AI models."

In all three cases, your conversation history survives. That's the standard Google is now meeting. The takeaway is simple: Gemini is moving from the most restrictive setup among major assistants to a mainstream one.

What should Gemini users do?

The practical advice is short. When the update lands in the coming weeks, open Gemini's settings and look for the new pair of controls. Decide separately how long you want your history kept, and whether you're comfortable with Google training its models on your text and attachments.

If you previously deleted your history to escape training, the old drawback is gone. Going forward, the two choices don't hostage each other.

It's also worth remembering that training settings across the industry default to "on" unless you act — Claude being the notable exception. Checking these toggles after every major AI update is a reasonable habit, because companies change defaults and redesign settings pages as products evolve.

The move signals where this is heading: as regulatory and legal scrutiny of AI training data grows, expect opt-out controls to get simpler and more independent across the industry. Google catching up to its rivals on this one setting suggests the era of bundling privacy choices together is ending.

Source: ZDNET AI

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

Staff writer covering regulation, policy and their impact on artificial intelligence and machine learning at AI In Context.

749 articles

Related articles

  1. Gemini’s Memory Summary Lets Users View and Edit What It Knows
  2. Google rolls out Gemini Personal Intelligence to Android Auto
  3. Google Gives Gemini Read-Only Access to Wallet Spending Data
  4. Google Kills Gemini's Gems and Migrates Them to 'Skills'
  5. Google DeepMind chief says Gemini 4 is nearly ready for launch

« Previous articleNext article »